Go-to Grok multi-agent QA

The desktop agent and HTTPS relay are free (MIT / hosted freemium). The mobile app is premium ($13 one-time) — spectator + veto, not required to run the loop. Independent Linespotting AB. Not affiliated with xAI.

Spec: USE-CASE-MULTI-AGENT-QA.md · skill: skills/qa/SKILL.md

Freemium agent + relay · premium app

Layer	Model	You get
`gbr-agent`	Free (MIT)	Pair, run, Bot API, fleet, `gbr_open` → `grok`, lock, result, tasks
HTTPS relay	Free (hosted) + MIT self-host	Mailbox through firewalls. No inbound ports. Optional self-host.
Mobile app	Premium $13 one-time	App Store / Play. Spectator + veto. Copies Bot API URL + key. Does not orchestrate.

Loop until solution

Grok Bot or gbr-mcp (orchestrator)     FREE
        │
        ▼
gbr-agent + free relay                 FREE
        │  gbr_open × 2
        ▼
grok writer TTY   +   grok tester TTY
        │
        ▼
Phone app (optional)                   PREMIUM spectator

Pin gbr-agent v0.6.0 — PINNED-INSTALL.md. gbr-agent run.
gbr_diagnose
gbr_open command=grok holder=writer and a second open holder=tester
Lock each session. Inject the fix into writer → gbr_result. Inject tests into tester → gbr_result.
gbr_tasks upsert until status=done. Release locks.

Do not share one session_id between two bots. Phone cannot inject (cannot_inject_phone). Hermes / OpenClaw / NemoClaw can be the orchestrator via stdio gbr-mcp — claw use case.